Computer Use MCP Server
Desktop control for AI models using TypeScript and nut.js integration.
Data last scanned today · Reviewed today
Overview
This server provides an MCP implementation allowing AI models to execute computer control actions locally. It mirrors the structure of official computer use guides while utilizing the nut.js library to handle mouse and keyboard inputs directly on your machine.
Our verdict
With a security score of 45/100, this server is classified as high-risk. The evaluation highlights that critical security aspects, including authentication methods and read-only mode support, have not yet been reviewed. Given that the tool grants an AI model full system control, it should be restricted to sandboxed environments under constant human supervision, similar to managing a hyperactive toddler, to prevent unintended system damage from model errors or prompt injections.
- Active development with commits within the last day
- Small supply-chain footprint with only five direct dependencies
- Open source MIT license
- High-risk potential due to full system control access
- Community-maintained without official vendor support
- Lacks reviewed authentication or read-only safety modes
Setup
{
"mcpServers": {
"computer-use-mcp": {
"args": [
"-y",
"computer-use-mcp"
],
"command": "npx"
}
}
}Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Some servers require API keys or tokens as environment variables — check the project's README and add an "env" object if needed.
- No read-only mode — running this grants full read/write access.
- No built-in authentication — anyone with access to this config can invoke every tool it exposes.
Tools
| Tool | Description | Risk |
|---|---|---|
| computer_use | Allows the model to control the computer by performing mouse and keyboard actions. | high |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›Are there any safety risks associated with using this MCP server?
Yes, as the server provides full control over your computer, models may make mistakes or fall victim to prompt injections. It is recommended to supervise the model closely and use this in a sandboxed user account.
›What is the recommended screen resolution for optimal performance?
Using a small, common resolution like 720p works best. If you cannot change your display resolution, try zooming in on active windows.
›How can I improve the reliability of web interactions?
Install and enable the Rango browser extension. This allows for keyboard-based website navigation, which is more reliable than coordinate-based clicking.
›Which AI models should I use with this server?
For best results, it is recommended to use the latest Claude models that are optimized for computer use.
›How is this server implemented?
The server implements tools near-identical to Anthropic's official computer use guide and utilizes nut.js to interact with your computer.
Changelog
Badge
Maintain this server? Add the live badge to your README.